In Olympic 500-meter speed skating, two skaters race simultaneously, one in the inner lane and one in the outer. The inner lane is shorter in circumference, so the skater who starts there crosses over to the outer lane halfway through. The question is whether the crossing — or anything else about the lane assignment — affects the time.
Hjort's bivariate mixed effects model on World Championship data showed a systematic disadvantage for the final inner-lane position of approximately 0.05 seconds. The effect was significantly different from zero — about three times its estimated standard deviation. In a sprint where hundredths of seconds separate medalists, 0.05 seconds is the difference between gold and nothing.
The unfairness parameter is too small to detect by watching individual races. Any single performance is dominated by the skater's ability, fitness, and form that day. The lane effect is buried in noise. It only emerges from the model, which separates the skater-level random effect from the lane-level fixed effect across many races.
The statistical analysis changed the sport. Starting with the 1998 Nagano Olympics, the format was reformed: skaters now race twice, once from each starting lane, with times averaged. The lane tax is zeroed out by design rather than by luck.
The broader pattern: a systematic bias that is invisible in any individual observation becomes undeniable across hundreds of observations. The fix doesn't remove the bias from any single race — it makes the bias symmetric, so it cancels.